strix-halo 开源项目推荐（267 个）

Osmantic /ODS

ODS V3 预发布版：在 V3 正式发布前进行公开测试与完善。将你的 PC、Mac 或 Linux 设备变成私有 AI 服务器。

Luce-Org /lucebox

面向异构硬件和消费级 GPU 的 LLM 推测推理服务器

peonist-ai /halogen-flash-server

像前沿 API 一样运行的本地推理：在 AMD Strix Halo 上运行 Qwen3.8-Flash-Next，并加上小型 NPU 模型，通过同一端口上的 OpenAI 和 Anthropic API 提供。

Soulmate-Halo /heterogeneous-gpu-pd-lab

使用稠密算力卡对稀疏大显存主机进行加速，主机包括 AI Max+ 395、M3 512G、DGX Spark 等。

gufo-org /gufo

The Open Source Strix Halo AI engine. Qwen 3.8 Flash Next Q4_K_XL: 1,801.01 pp, 73.17 tg single user, 194.37 tok/s 8 users; Qwen 3.8 27B Q4_K_XL: 656.33pp, 70.56tg tok/s single user with DFlash2

hogeheer499-commits /strix-halo-guide

AMD Strix Halo本地LLM指南：为Ryzen AI MAX+ 395 / Radeon 8060S、Ollama、llama.cpp Vulkan/RADV、ROCm、原始证据、直接100 t/s 30B Qwen、140 t/s CHADROCK MTP、120B GGUF。

julianmb /q38rocm

Qwen 3.8 27B ROCmFP4 on AMD Strix Halo (Ryzen AI Max+ 395). Up to 36 tok/s via MTP Speculation, TurboQuant & Mesa RADV Wave64.

Yamz-Labs /kyojin

Kyojin: open-source engine that runs 100 GB-class open models (GLM-5.3, Qwen3.8, DeepSeek V4.1) on one desktop machine. AMD Strix Halo, NVIDIA RTX and Apple Mac. OpenAI API, built on ExLlamaV3.

Nathanw1014 /strix-halo-llamacpp

Performance-tuned llama.cpp for AMD Strix Halo (gfx1151): FA + MoE-prefill fixes with a bundled current Mesa driver. Vulkan and HIP; portable dir, Docker, and distrobox.

peonist-ai /halogen-server

The fastest way to run Qwen3.8 27B on Strix Halo (gfx1151)

IIIIIllllIIIIIlllll /Qwenox

An inference framework customized for AI MAX 395, compatible with both Linux and Windows, and supports deploying only qwen3.8-flash-next. This project is for reference and learning purposes only. 为AI MAX 395定制的推理框架，Linux和Windows通用，只能部署qwen3.8-flash-next。本项目仅供参考学习使用。

LaurentZuijdwijk /llama.cpp

llama.cpp with adaptive speculative decoding (--spec-draft-adaptive) and a Vulkan backend tuned for AMD Strix Halo. 4.7x on structured output, 1.9x mainline prefill on MoE.

shawnshekari /strixite

LLM inference engine written from scratch in HIP for AMD Strix Halo (gfx1151) and Qwen3.8-Flash-Next - ~50 tokens/s out to 492k context

julianmb /halofpx

High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-edge RDNA 3.5 Wave64/ROCmFP4 kernels, and silicon-tuned profiles.

mikealanni /gufo

GPU + XDNA2 NPU inference for AMD Strix Halo (gfx1151): Qwen3.6-35B-A3B, Qwen3.8-27B, Qwen3.8 Flash-Next, Flint and GLM-5.3 Flash (321B MoE). NPU prefill up to +28%, Flint 4.1k tok/s prefill, MTP / DFlash2 / auto speculative decoding up to 159 tok/s on structured text.

Hal0ai /hal0

Open-source self-hosted home AI inference platform for AMD Strix Halo — multi-backend slots, OpenAI-compatible gateway, Vue 3 + FastAPI + systemd.

deseven /strixhalo-homelab

This is a mirror of the Strix Halo HomeLab wiki, to browse the wiki click on the link below

rulith-dev /rulith-inference

Rulith Inference, formerly Strix Llama. Serving a 125B MoE model fast on one AMD Strix Halo machine, on Windows: a patched llama.cpp, a manager, and a desktop app.

ndom91 /ha-voice-rocm

Experimental support for many TTS/STT LLMs wrapped in a Wyoming API for consumption via Homeassistant

hec-ovi /text-to-3D-skill

Local text to textured GLB on an AMD Strix Halo iGPU (gfx1151): FLUX.2 klein, a Vulkan-only TRELLIS.2 engine, and humanoid auto-rigging with SkinTokens on ROCm. No Blender, no CUDA.